The brain is a computer, not a philosophy — and a business route around drugs
Max Hodak's judgment: wiring up the brain as a computer gets you effect sizes and iterability that small-molecule drugs cannot reach — the Prima retinal prosthesis is already approved for sale in Europe, but the field of view is like looking through a straw, and it is black-and-white only.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Restoring sight isn't the mission; it's the cash flow that funds it
Science was founded with two tracks. One was a biohybrid neural interface: no metal electrodes, no genetic modification, but transplanting living neurons in and letting them grow new biological connections. That is a large research program, and it had to be paired with a business that could make money in the near term. So they asked themselves what the most valuable thing to do was, given the resources they had and the state of the field in early 2021, and the answer was restoring sight to the blind. That structure has determined every trade-off since: vision is not the mission itself, it is the cash flow that funds the mission.
— Max HodakIf the optic nerve is intact, the work belongs at the retina
There are three places to inject signal if you want to restore vision: the retina, the lateral geniculate nucleus in the thalamus, or V1 -- that half-billion cells at the back of the brain. Hodak says a pile of scientific and technical reasons all point to the same conclusion: as long as the patient still has an optic nerve, you should do it at the retina. They worked through the variants -- they developed their own gene therapy aimed at the retina (which may go into humans next year), and they surveyed both electrical stimulators and ultrasound. The conclusion was that the best work in the world on the electrical stimulation path was not in their own hands.
— Max HodakThe vision breakthrough came from an acquisition, not from the lab
The best electrical-stimulation retinal approach belonged to a small French company, Pixium: the technology came from an inventor at Stanford, was licensed to Pixium, and by the end of 2022 was already in clinical trials. Science had known them for a few years, and then bought the company at the point where it saw something that essentially nobody else was seeing at the time. After the acquisition it took roughly two more years to push the technology to an approvable state; European marketing approval came in July, and the first sales followed in the weeks after that. This is the most operational item in the episode: the breakthrough came from valuation arbitrage plus two years of engineering cleanup, not from a flash of insight in a lab.
— Max HodakDevices aren't better than drugs; they let you keep iterating after failure
Hodak's argument is not that devices beat drugs, but that the failure structure is different. In drug development you work for ten years, run the trials all the way through, turn the card over and it may say no -- and then everyone goes home. On the brain-interface path, by contrast, in his words they have a clear sense of how to make the thing better. He even brings in highly engineered patient-specific CAR-T for comparison: what you may get in exchange there is an enormous immune overreaction. At the same time he does not gloss over where Prima stands today. This is the first time anyone has restored form vision in a blind person's mind's eye, but the field of view is very small, like looking through a straw, and it is black-and-white only. The next step is adding grayscale depth; there is a path to seeing red and green, while blue is harder.
— Max HodakScience is run like a technology company, not a biotech company
Hodak admits his position runs against the current in the industry: BCI is being widened out beyond motor decoding, and the seemingly hair-splitting question of whether a retinal prosthesis counts as a BCI matters, because once the answer is yes, it opens up a large stretch of medicine that nobody had thought about that way. He puts this down to culture: biotech has always had an East Coast / West Coast divide, and Science is more like a technology company than a traditional biotech company -- looking at old biology problems through a device lens tilts it further toward tech. The result shows up in the financing structure: the money comes mainly from technology investors, and in the Series A the only biotech VC he actively went out and sought was Bob Nelson.
— Max HodakModel and brain representations align closely enough to use as a tool
He hangs the rationality of investors pouring into BCI on the platonic representation hypothesis: look inside large models and the mathematical objects there resemble what you see in neuroscience -- the way a model represents a concept and the part of the brain that represents that concept are geometrically very similar. The key point is that he says this is not a bystander's observation; Science is using it constructively. They can align neural recordings from animal brains with the internal representations of AI models. For him the significance runs past neuroscience: it is evidence that AI is not hitting a wall, that there is something deeper and real here. He also names the contested side -- whether this structure holds globally or only locally is not settled within the field, and there is a camp that simply does not want it to be true.
— Max HodakA prosthesis showing only light blobs still cost $150,000 per patient
Pricing has not been set and he deliberately says little about it, but he does lay out the anchors. Roughly ten years ago, Second Sight's retinal prosthesis could only produce flickering blobs of light that patients had to assemble into meaning themselves -- it did not get you form vision -- and yet each patient paid about $150,000. And a gene therapy that is meaningful for only about 5% of patients inside a single narrow indication, delivers 0.1 lines of improvement, and essentially just slows the rate of degeneration, is reimbursed at close to $500,000 per eye. His explanation has two parts: some of it is that these therapies are expensive to develop and have historically failed at a high rate, and some of it is that vision matters so much that losing it is completely disabling. On addressable population, the current version covers on the order of hundreds of thousands of patients in the US and Europe, and the next generation (in animal trials now, with humans possible next year) would expand that to millions.
— Max HodakEvery organ other than the brain is a supporting actor
The core of his worldview: the only organ that cannot even in principle be transplanted is the brain. The heart, the pancreas, the liver and the lungs are, as he sees it, supporting actors, and their reason for existing is to keep the brain's activity going and to keep it interesting. The skull is the vat -- the brain connects to its environment through very few wires, the cranial and spinal nerves; the optic nerve is the second cranial nerve, and the vestibulocochlear nerve, which carries hearing and balance, is the eighth. So if you can move vision, hearing, balance, somatosensation and motor signals into and out of the brain, that in itself is the endpoint. He points out that the two leading causes of death are cardiovascular disease and cancer that has metastasized to the brain, and argues both can be attacked along this path; neurodegeneration remains very hard. And if humans really are going interstellar, they will have to adapt to hard vacuum -- which means preserving yourself and upgrading yourself turn out in the end to be the same project: substrate independence.
— Max HodakWiring your brain to AI won't be faster; the bottleneck is 10 bit/s
He explicitly rules out the easiest BCI startup story to tell right now -- non-invasive, high-bandwidth conversation with AI models. His reasoning has two layers. First, speaking or writing is itself thinking; the intuition that there is a fully formed thing in your head and it would be faster if only a BCI could read it out is misleading, because you discover it was not formed the moment you actually sit down to write. Second, cognitive bandwidth already has a deeply evolved bottleneck of about 10 bit/s -- fly someone with an exceptional memory over Manhattan in a helicopter and have them draw what they saw, and an hour or two of drawing works out to roughly that number, and there are several independent lines of evidence pointing the same way. He allows that wearing a cap and querying an AI with your inner monologue as you walk around may well be feasible (some combination of EEG and MEG), but it may turn out like AR glasses: attention is already 100% occupied, and moving it onto your face does not change anything.
— Max HodakIn their own words · checked verbatim
It's like if I put electrodes in M1, you will probably be using a computer in an hour. … You can do drug discovery for a decade, run a clinical trial, you're going to turn over a card, the answer might be no
Max Hodak8:04
Well, if you want to make people angry, you should tell the internet that the brain is a computer.
Max Hodak10:04
if you can if you get vision, hearing, balance, and a kilobit per second of motor control, you're halfway to the Matrix.
Max Hodak14:05
we can get alignments between animal brain neural recordings and AI model internal representations. And that was a big clue to me that the AI … was on the right track and this was like not a gimmick and not hitting a wall.
Max Hodak20:10
And so, I'm going to be ultimately fairly disappointed if I'm murdered by my pancreas.
Max Hodak26:15
Figures
| Prima European marketing approval | Approved in July, with the first sales in the weeks that followed | 5:03 |
| Time from the Pixium acquisition to approval | About 2 years | 5:03 |
| That gene therapy's effect and coverage | 0.1 lines of visual acuity improvement, only about 5% of patients, a single narrow indication | 24:14 |
| AMD prevalence | By age 80, about one in two people show early changes; at 85, about one in ten is diagnosed | 22:12 |
| Prima addressable patient population | Hundreds of thousands in the US and Europe for the current version; millions for the next generation | 24:14 |
| Cognitive bandwidth bottleneck | About 10 bit/s | 29:16 |
Glossary
- form vision
- Seeing a coherent face or paragraph that your eyes can scan across, rather than scattered blobs of light.
- retinal prosthesis
- A chip implanted under the retina that bypasses dead rods and cones and stimulates the retina directly.
- biohybrid neural interface
- An interface approach that transplants living neurons to grow new biological connections, in place of metal electrodes.
- platonic representation hypothesis
- Different models and brains converge, as they learn, on the same underlying representational structure.
- connectomics
- Mapping the full set of neural connections in a brain; close for mice, still far away for humans.
- substrate independence
- The mind does not depend on one particular physical substrate and can be moved onto another.
How to listen
Investors in medical devices and hard tech, founders building BCI or neurotechnology, and engineers who care about how AI representations relate to brain representations.
16:07-19:10, the speculation about uploading and the continuity of consciousness, can be skipped.